Fri, Sep 18, 2015 at 02:39:10PM CEST, fbl(a)sysclose.org wrote:
On Fri, Sep 18, 2015 at 10:05:54AM +0200, Jiri Pirko wrote:
> Fri, Sep 18, 2015 at 12:00:51AM CEST, sridhar.samudrala(a)intel.com wrote:
> >
> >
> >On 9/17/2015 2:10 PM, Flavio Leitner wrote:
> >>On Thu, Sep 17, 2015 at 11:45:20AM -0700, Samudrala, Sridhar wrote:
> >>>On 9/17/2015 12:42 AM, Jiri Pirko wrote:
> >>>>Can you please provide some more info about this change, like when
in
> >>>>the spec this is decribed? Thanks!
> >>>Section 5.3.15 in 802.1ax spec defines these flags
> >>>http://varlyakov.com/diplomna/diplomna/802.1ax/802.1ax.pdf
> >>>Here is another link that describes the LACP flags.
>
>>>http://networkingbodges.blogspot.com/2012/12/all-sorts-of-things-about-lacp-and-lags.html
> >>Are you actually seeing a bug without this patch?
> >>
> >>COLLECTING and DISTRIBUTING are informative bits about the port's
> >>state, but they are secondary to other states like SYNCHRONIZATION,
> >>so I don't see how that can cause an issue.
>
> IIRC, I originally decided to set those 2 bits always, because there is
> not an actual state machine inside teamd_lacp as it is described in the
> standard, so the internal status does not make a lot of sense.
>
> >>With the proposed patch applied, there could be an outage if
> >>the state moves to EXPIRED or DEFAULTED.
> >With the current code, these bits are always set even if the partner is not
> >present. Is this OK?
> >When the state moves to EXPIRED or DEFAULTED, isn't it correct to mark these
> >bits as FALSE?
> >It looks like teamd implementation is not really looking at these bits in the
> >received LACPDUs.
> >
> >This is more of an issue detected by our validation team looking at the LACP
> >frames generated by
> >teamd rather than an actual interoperability issue with other
> >implementations.
>
>
> If it does not break interoperability, I would leave this as it is.
> Unless there is an argument againt it. Thoughts?
He has a valid point. Teamd shouldn't push frames if the other side
hasn't set COLLECTING bit otherwise there might be packet loss.
Also, teamd shouldn't tell the other side to send when it isn't ready.
But as it is today, we might have packet loss but still the packets
are flowing. With the patch applied, we have less issues with packet
loss but there is no connectivity in the other states.
To follow the standard and fix that and other issues, teamd needs the
MUX state machine which relies on selected and sync states.
fbl
Fine, so you are ok with the patch as it is?
> >>So, without more details for this change, NACK.
> >>
> >>fbl
> >>
> >>>Thanks
> >>>Sridhar
> >>>
> >>>>
> >>>>Thu, Sep 17, 2015 at 12:47:36AM CEST, sridhar.samudrala(a)intel.com
wrote:
> >>>>>Signed-off-by: Sridhar Samudrala
<sridhar.samudrala(a)intel.com>
> >>>>>---
> >>>>>teamd/teamd_runner_lacp.c | 3 ++-
> >>>>>1 file changed, 2 insertions(+), 1 deletion(-)
> >>>>>
> >>>>>diff --git a/teamd/teamd_runner_lacp.c
b/teamd/teamd_runner_lacp.c
> >>>>>index 09899ce..f663016 100644
> >>>>>--- a/teamd/teamd_runner_lacp.c
> >>>>>+++ b/teamd/teamd_runner_lacp.c
> >>>>>@@ -900,7 +900,8 @@ static int lacp_port_actor_update(struct
lacp_port *lacp_port)
> >>>>> state |= INFO_STATE_LACP_TIMEOUT;
> >>>>> if (lacp_port_selected(lacp_port))
> >>>>> state |= INFO_STATE_SYNCHRONIZATION;
> >>>>>- state |= INFO_STATE_COLLECTING | INFO_STATE_DISTRIBUTING;
> >>>>>+ if (lacp_port->state == PORT_STATE_CURRENT)
> >>>>>+ state |= INFO_STATE_COLLECTING | INFO_STATE_DISTRIBUTING;
> >>>>> if (lacp_port->state == PORT_STATE_EXPIRED)
> >>>>> state |= INFO_STATE_EXPIRED;
> >>>>> if (lacp_port->state == PORT_STATE_DEFAULTED)
> >>>>>--
> >>>>>2.1.0
> >>>>>
> >>>_______________________________________________
> >>>libteam mailing list
> >>>libteam(a)lists.fedorahosted.org
> >>>https://lists.fedorahosted.org/mailman/listinfo/libteam
> >
> >_______________________________________________
> >libteam mailing list
> >libteam(a)lists.fedorahosted.org
> >https://lists.fedorahosted.org/mailman/listinfo/libteam
>